A Pathomics-Based Prognostic Model for Disease-Free Survival in Resected Gastric Cancer.

Area of Science:

• Oncology
• Medical Imaging
• Biostatistics

Background:

• Gastric cancer (GC) prognosis remains challenging, necessitating improved predictive tools.
• Integrating quantitative imaging features (pathomics) with clinical data may enhance prognostic accuracy.

Purpose of the Study:

• To develop and validate a prognostic risk model for disease-free survival (DFS) in gastric cancer (GC) patients.
• To integrate pathomics features from pathological images with clinical variables for improved prediction.

Main Methods:

• Retrospective enrollment of 393 GC patients into training (n=275) and validation (n=118) cohorts.
• Extraction of pathomics features from pathological images; LASSO-Cox regression for feature selection.
• Development of a clinic-pathomics model and nomogram for DFS prediction, validated using time-dependent ROC analysis and decision curve analysis.

Main Results:

• 16 pathomics features were selected; a high-risk group showed significantly worse DFS (HR > 2.26, p < 0.0001).
• The clinic-pathomics model achieved strong predictive performance (AUCs up to 0.851 in training, 0.702 in validation).
• The nomogram demonstrated high calibration and outperformed clinic-only and pathomics-only models.

Conclusions:

• A clinic-pathomics model integrating pathomics and clinical data reliably predicts DFS in GC patients.
• This integrated approach facilitates individualized DFS predictions and personalized treatment strategies.